{
  "id": 19146628,
  "title": "Human brains track who did what to whom in ways AI language models don’t, fMRI study suggests",
  "url": "https://urgent.news/2026/11/09/human-brains-track-who-did-what-to-whom-in-ways-ai-language-models",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-09T11:00:24.000Z",
  "source": {
    "name": "PsyPost",
    "slug": "psypost",
    "url": "https://www.psypost.org/human-brains-track-who-did-what-to-whom-in-ways-ai-language-models-dont-fmri-study-suggests/"
  },
  "original_language": "en",
  "account": "A recent neuroimaging study published in eLife suggests that human brains process sentence meaning differently from large language models, such as those used in AI chatbots like ChatGPT. The study found that when sentence meaning relies on word order rather than vocabulary alone, brain activity aligns more closely with computational models that track grammatical roles, like who is performing an action and who is receiving it. This semantic tracking is achieved through specific brain regions, specifically the left temporal cortex. In contrast, large language models organize meaning using transformers that represent word and sentence meanings as vectors. Researchers conducted a study involving 108 single-clause English sentences, divided into six subsets, to isolate sentence structure from vocabulary. These sentences were designed with role-swapped nouns while keeping the words nearly identical, making it difficult for a model to distinguish them based on vocabulary alone. The study evaluated four modeling approaches, with two using numerical vectors and two explicitly encoding structure. Thirty native English speakers, aged 18 to 40, read the sentences in an fMRI scanner and answered comprehension questions. The researchers analyzed brain activity from the middle three seconds of each trial, and similar results were found in the final window.",
  "summary": "Large language models can sound remarkably human, but a brain imaging study suggests they organize sentence meaning differently from people when word order changes the meaning, as in “the dog bit the man” versus “the man bit the dog.”",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}